China leverages subsidized open-source AI models for global ecosystem lock-in while US firms pivot to proprietary systems

Chinese Open Source AI companies capture virtually no revenue and the ratios vs US companies is insane

Zhipu / http://Z.ai, the company behind GLM 5.2, is at a $137B market cap on about $107M of FY25 revenue

That means the stock trades at roughly 1,280x FY25 sales

For the stock to trade at 50x sales, revenue needs to reach ~$2.7B/year, about 26x current FY25 revenue. At 20x sales, it needs ~$6.9B/year, about 65x current FY25 revenue

Side note I love GLM 5.2 it’s insane

MiniMax is another one: ~$23B market cap on ~$79M FY25 revenue = ~290x sales

Alibaba / Qwen is at ~$245B market cap on ~$151B FY26 revenue = ~1.6x sales, the lowest P/S of this group but not pure AI

Now compare that to the U.S. labs:

- OpenAI ~$25B annualized revenue and ~$852B last private valuation = ~34x sales

- Anthropic ~$47B annualized revenue. At the reported ~$965B valuation, it’s ~21x sales

Chinese multiples need to come down (in the form of more revenue if not the market caps are hard to sustain) and US ones probably go up as revenue shifts slightly to substitutes and valuations go higher post IPO on thin floats Chinese companies cede a ton of revenue to inference providers (OpenRouter, Venice, BaseTen, others) since people want these models but don’t want to send data to China

China’s model companies need to somehow show they aren’t retaining data and undercut everyone on pricing if they want their API revenues to inflect. Seems very hard to do culturally/socially

Another option is for Chinese AI companies to own part of the U.S. inference providers and do deals to pre-release the top models to them first for a cut of the revenue maybe

I.e. you get GLM 6 on popular inference providers and Zhipu gets a big cut. Money flows to the Chinese model cos to the detriment of inference providers (% wise) but the pie gets a lot bigger